Apple’s Siri Update Could Include a Major AI Privacy Twist

More rumors about Siri’s big makeover are leaking ahead of Apple’s annual developer conference—and one big change could have a lot to do with data privacy and security. Apple is expected to launch a standalone app for its embattled AI assistant, Siri, which will operate as a chatbot-like interface, similar to Anthropic’s Claude or OpenAI’s ChatGPT. Users are expected to be able to type or speak requests. Although the app will be able to store conversations to mine for contextual information for future requests, Apple is expected to differentiate itself from its competitors by allowing users to auto-delete their conversations, according to a report from Bloomberg. This is much like the way Apple allows users to auto-delete their text messages, which works so well, Bloomberg notes, that it has invited complaints over government officials using the feature to delete their messaging histories. Apple also plans to establish stricter guidelines around what information does hang around in the system and how long it can be kept. Competitors typically allow users to toggle on temporary incognito modes that prevent conversations from being used to train AI models. It’s worth noting that Apple has brought in Google’s Gemini and cloud infrastructure to keep the Siri update functioning and on schedule—after costly delays. Apple first introduced its Apple Intelligence as a mix of on-device and cloud-based computing that it billed as similar to iPhone security but in the cloud. Called Private Cloud Compute, it was expected to operate on Apple’s own servers and chips. Bloomberg reports that Apple will still call its system “Private Cloud Compute,” but the mechanics of how it might operate aren’t clear, given the Google integration. The two-year delay around the roll out of an updated Siri has proved to be a stain on Apple’s recent track record, as well as an expense, because of a recent $250 million settlement on a class-action suit that alleged false advertising. But if Apple is successful in relaunching Siri with an emphasis on privacy, it could justify to consumers the longer wait time, especially as concerns gather about what people are sacrificing in the name of cutting-edge AI features. Apple’s Worldwide Developers Conference will kick off on June 8. The tech giant is expected to announce iOS 27, as well as software for Mac computers and iPads. And the graphic for the conference, featuring glowing lettering, hints at a new look and functionality for Siri. BY CHLOE AIELLO @CHLOBO_ILO

They Won a Prestigious Writing Prize. Then These Key Giveaways Sparked Allegations of AI

A London-based literary competition is facing major scrutiny after three of five winners have been accused of using AI—partly or wholly—to write their prize-winning stories. The 2026 Commonwealth Short Story Prize selected one winner each from five regions that span Africa, Asia, Canada and Europe, the Caribbean, and the Pacific. Following publication of the winning entries in literary magazine Granta, online sleuths called foul. The Caribbean regional winner, Jamir Nazir, was praised for the “lyrical precision and haunting atmosphere” of his short story “The Serpent in the Grove,” as well as “the confidence and restraint of its voice,” according to a post on social media platform X by Commonwealth Foundation Creatives. But internet denizens allege that the very same voice that won the prize may not be human at all. Nabeel S. Qureshi, an AI marketing entrepreneur and a former visiting scholar at the Mercatus Center at George Mason University, flagged certain signs he said were AI tells such as “‘Not X, not Y, but Z’ sentences,” as well as the use of the word “‘hum,'” in a post on X. He concluded by writing: “A major milestone for AI, at any rate …” Following the allegations, Wired ran the text of “The Serpent in the Grove” through AI detection tool Pangram, which the publication notes has consistently outperformed other similar tools. It determined that the text was 100 percent AI-generated. It’s worth noting that no AI-detection tools are totally accurate. “The Serpent in the Grove” isn’t the only story under scrutiny. Pangram determined that “The Bastion’s Shadow” by John Edward DeMicoli, the winner from Malta, was also fully AI-generated and that “Mehendi Nights” by Sharon Aruparayil, the winner from India, was partially written using AI. Holly Ann Miller’s “Second Skin,” and Lisa-Anne Julien’s “Me and Ma’am,” however, were ruled “fully human-written” by Pangram, according to Wired. The regional winners were chosen from 7,806 entries, which the Commonwealth Foundation noted on its site is the “second highest number in the Prize’s history.” A final winner from among the five will be announced on June 30. Following the allegations, the Commonwealth Foundation released a statement on its website, acknowledging the challenges generative AI poses to literary and creative work. The statement also noted that the foundation’s judging process is robust, but judges do not currently use AI checkers in any stage. “We are aware of allegations and discussion regarding generative AI and our Short Story Prize. We take these claims seriously and are committed to responding to them with care and transparency,” the statement reads. “When they submit stories to the Prize, writers accept our entry rules and guidelines. These include confirming that their submission is their own original work. All shortlisted writers have personally stated that no AI was used and, upon further consultation, the Foundation has confirmed this.” The foundation also noted that until a reliable tool emerges with which the organization can screen unpublished work for AI, the prize competition “must operate on the principle of trust.” As always, Redditers had much to say about the subject, some assuming AI guilt, others questioning the accuracy of the AI checking tools, and many picking at the quality of the stories more broadly. “This…doesn’t surprise me given the state of contemporary literary prose. It honestly just reads like bog-standard ‘MFA voice,’” one Redditor wrote of Nazir’s story. A recent report from digital marketing agency Graphite found that since the debut of ChatGPT in 2022, there has been a meteoric rise of AI-generated content on the internet. The number of articles written by AI now equals that of human-written content, although the overall share seems to have plateaued. Axios reported at the time that the quality of AI-generated writing has meaningfully improved, not to mention that it can be difficult to determine what constitutes AI writing. Whereas some content is mostly or entirely AI-generated, some writers use AI tools throughout the process of drafting and editing. BY CHLOE AIELLO @CHLOBO_ILO

AI is changing the internet forever. Here’s how

There’s a simple reason Google is making sweeping changes to its iconic, decades-old search engine: users are making complicated requests. “People are asking much longer and harder questions that no longer have a clear response anywhere on the internet,” said Robby Stein, vice president of product for Google Search. Stein spoke to CNN about a new feature that lets Google generate custom visuals, interactive graphics and even mini-apps running on Google’s search page in response to queries by piecing together sources from across the web. It’s one of many updates the internet giant announced at its annual conference this week. The most valuable real estate on the internet is evolving to reflect the new ways people find information online, the latest example of how artificial intelligence is changing the internet across search, social media, online shopping and more. People are starting to use longer, more specific search terms instead of succinct generic keywords, according to Google, and are increasingly beginning their searches in apps like ChatGPT, experts say. Fake, AI-generated influencers are causing a stir on social media. And people are increasingly using AI to compare and buy products. It’s getting impossible to avoid using the internet without somehow encountering AI, despite growing anxiety about the tech and its impact on jobs, safety and the environment. “After a while, it just becomes part of the way you live,” said Joseph Turow, a University of Pennsylvania media professor who will soon be releasing a book about AI’s impact on internet advertising. ChatGPT ‘trained’ people to search differently Google says its search box is getting its biggest upgrade in 25 years. The new search field expands to fit more text and makes it easier to add other media to a search — like photos, files and Chrome browser tabs. The goal is to shrink the number of steps for a user to complete a search, according to Stein. That includes tasks like performing a search based on a photo or switching to Google’s AI Mode before asking a follow-up question. Searches that involve questions based on snapping a photo or circling something on a phone screen are growing 60%, year-over-year, he said. Searches in AI Mode, or the version of Google tailored for back-and-forth interactions, have more than doubled every quarter since they launched a year ago, and AI Mode queries are triple the length of a regular search on average. Data from SEO and marketing firm Semrush indicates some people are starting to search Google the way they type to ChatGPT. Searches containing 11 words or more increased from 3.27% to 5.37%, and conversational queries jumped from 5% to 20%, while keyword-style searches decreased. Yet the median query still contains just three words, suggesting that most people still search the old-fashioned way. Robert Langenback, president of SEO marketing agency Eight Oh Two Marketing, said he’s observed people typing in more searches that range from three to five or five to 10 words instead of two to three words. That started before ChatGPT’s arrival in late 2022, although it’s ramped up significantly since then. “(AI has) really almost trained people how to search differently,” he said. People generally use a mix of AI apps like ChatGPT and Google. More than 20% of ChatGPT referral traffic goes to Google, Semrush found after analyzing 1 billion lines of US clickstream data, or “trails” of user activity across the web. Google is typically used for direct questions or transactions, while ChatGPT is used for summarizing information, making comparisons and drafting materials, Semrush said in an email to CNN. “There’s a lot of just, ‘I’m trying to find something and help me get to it right away,’ that is the bulk of the queries that have gone into Google over time,” said Leigh McKenzie, director of organic visibility at Semrush. The rise of AI influencers AI’s reach extends far beyond search. Take Aitana Lopez’s Instagram profile. Online she looks like any other social media influencer, photos showing her posing at glitzy events, hitting the gym and sharing beauty tips to nearly 400,000 followers. But she’s not real. Lopez is one of the most prominent AI-generated characters to rise to internet stardom, along with Lil’ Miquela, Lu do Magalu and Granny Spills. Nearly 80% of marketers have increased spending on creator content that uses generative AI in the last 12 months, according to social agency Billion Dollar Boy. There are even awards celebrating the best AI-generated internet personalities. AI personalities are appealing to brands because they’re typically cheaper than high-profile human influencers and can morph to fit specific campaigns, said Turow. Tech giants want to make AI an even bigger part of social media. Meta is integrating its Muse Spark model into apps like WhatsApp, Instagram and Facebook and is testing side chats with its AI assistant in group conversations. On Tuesday, Google announced Gemini Omni, a new AI model that people can use to generate realistic avatars of themselves. The race to own online shopping Traffic to US retail sites from AI services grew 393% year-over-year in the first three months of 2026, according to Adobe, with Meta, Amazon, Google and OpenAI all introducing AI shopping tools. Google this week introduced a new “universal” shopping cart that allows users to add items from different retailers across the web. Amazon recently folded its Rufus shopping assistant into a new tool called Alexa for Shopping, which incorporates the AI helper into the online retailer’s search bar so shoppers can ask it to compare products and pricing history, among other things. But even as AI directly answers shoppers’ questions at the top of Google, Stein says there’s still a need for quality websites created and maintained by humans. Google says it still send billions of clicks to websites every day, although Pew Research data last year found that Google users are less likely to click links when viewing an AI summarized answer. Langenback says that while his clients are seeing less traffic, the traffic they are getting is leading to higher engagement — completing a purchase, booking an appointment or requesting a quote. “You just have to be ready to adapt, because (search) could look a lot different six months or a year from now,” he said. By Lisa Eadicicco

Google is making its biggest change to the search bar in years

To get ahead in the new internet age, Google wants to help you google less. The company on Tuesday revealed a flurry of AI-powered features for its search engine, AI assistant Gemini and other services. It’s part of Google’s latest effort to revamp its decades-old business model to fit the era of artificial intelligence. Among those updates is a new version of the search bar that can crawl the web on a user’s behalf and a new mode in Gemini that can work autonomously over periods of time. The changes bring Google’s search engine closer to the likes of its biggest competitors today: Anthropic and OpenAI, whose sophisticated AI models have taken over some of the duties of search tools and web browsers. Revamped search Google for years has been moving away from delivering a list of blue links in response to search queries. But the refreshed search engine, which runs on the company’s new Gemini 3.5 Flash model, represents what may be its biggest shift yet toward AI and away from traditional search. The new search field expands to accommodate longer queries that are more conversational, aligning with the way one might type or talk into Gemini or ChatGPT. Users will be able to create “agents” in Google’s search engine that can track or research topics on their own. Google says it’s useful for tasks that require tracking and monitoring announcements and listings over time, like apartment hunting or new apparel releases. One can, for example, enter a query like “Keep me updated when any of my favorite athletes announce sneaker collabs or signature drops” to prompt Google to monitor announcements from notable athletes and brands, the company cited as an example in a press release. Google will also now generate custom visuals and mini apps in response to certain requests, such as creating a fitness tracker that incorporates a person’s location, weather data and apps connected to their Google account. A new Spark Since it launched the AI-powered Gemini, Google has struggled to differentiate the assistant from its main search engine. Spark, a new mode within Gemini that can work on tasks in the background, is its latest attempt to change that. Spark will be able to work on recurring long-term tasks like monitoring credit card statements and email inboxes for important updates and creating summaries or to-do lists. It can also reference content across certain apps, like compiling notes from Google Docs, Gmail and Slides, and the company says more third-party apps will be supported in the future. The company is also adding Spark to the Gemini app on Mac computers so that it can work with local files, and users will be able to monitor what their agent from their phones through a new feature called Android Halo. The agent will stay active even when the person’s laptop is closed or their phone is locked, Google says. The focus on autonomous features seems like a direct response to OpenClaw, the buzzy AI agent that made waves in Silicon Valley earlier this year for its ability to run programs and commands without constant prompting from the user. Building AGI Google has been pursuing AI agents for years, although use cases have mostly focused on specific tasks like shopping or email management and haven’t taken off with consumers broadly. That’s largely because the technology simply hasn’t been reliable enough. “I think there’s this uncanny valley where the models aren’t yet good enough, so you can’t trust them fully, and so you aren’t really sure what you can and cannot do,” Tulsee Doshi, senior director of product management at Google DeepMind, told CNN. Google hopes the updates will bring it closer to its big-picture goal of developing artificial general intelligence: a theoretical stage of AI in which the technology becomes as intelligent as a human at broad range of subjects. OpenAI, Meta and others are racing to be the first to get there. But AI will have to get better at updating its own intelligence before AGI is possible, said Koray Kavukcuoglu, chief technology officer at Google’s DeepMind AI lab and the company’s chief AI architect. “Right now, our models (have) some sort of capability in doing that, but you can imagine that they’re a little bit static in time,” he told CNN ahead of Google’s conference to announce the updates. DeepMind DeepMind is at the center of the company’s AI strategy and has become one of its biggest assets in the AI race. It’s Google’s “secret weapon in the AI wars” according to Dave McCarthy, an analyst covering cloud and infrastructure services for market research firm The International Data Corporation. Most tech companies don’t have massive consumer reach and direct access to a research lab and cloud systems. “Google is the only company that I can think of that actually has a play in every one of those areas,” McCarthy said. Yet Anthropic and OpenAI are largely perceived as being ahead of Google in AI business products; Anthropic has been releasing new models and AI agents for coding, finance and other office work at a rapid clip this year. Anthropic accounted for 34.4% of paid AI business subscriptions in the US in April while OpenAI accounted for 32.3% and Google’s share was just 4.5%. That’s according to finance platform Ramp, which analyzed contract and transaction data with AI companies from more than 50,000 American businesses. AI is also causing concerns over the future of jobs, safety and the impact of data center construction on local communities and the environment. Half of American adults say the increased use of AI in everyday life makes them feel more concerned than excited, according to Pew Research. But Google, like many companies, is staking its future on the technology. Gemini now has more than 900 million active users, and the company expects to spend about $180 to $190 billion this year on expenses related to AI infrastructure and chips, Alphabet CEO Sundar Pichai said in a press briefing ahead of the conference. And the technology will undoubtedly continue to move quicky. Varun Mohan, a director at Google DeepMind who works on Google’s Antigravity AI coding product, said they ship a new release “close to every day” for internal developers. “We’re open to the fact that we are going to need to make changes very quickly, because otherwise we are going to have a product that is old for our users,” he said. “And we’ll be doing our users a disservice if we just hold on to our ideals of what the product is today.” By Lisa Eadicicco

AI Tools Are Rewriting Business Security, and Not in a Good Way

AI is completely rewriting the script on how founders run their businesses. As founders implement more AI tools into their workflows, they need to understand the security of their AI software supply chain. Only recently, deployment platform Vercel suffered a massive security breach as the result of an employee connecting a third-party AI tool to their corporate Google account. Revolutionizing your business operations isn’t going to do much good if sensitive data is compromised. Before you roll out AI-powered tools, you must consider how they affect the entire software supply chain. What are the risks of an under-managed AI software supply chain? Currently, enterprise enthusiasm for AI adoption seems to be outpacing companies’ ability to enact meaningful security measures. According to a report by cloud and AI security solutions provider Wiz, while 87% of security professionals are using some type of AI service, only 13% have an AI-specific posture management security strategy. Twenty percent aren’t implementing any type of AI security strategy. Another 25% admit they don’t know which AI services are currently being used in their organization. The lack of information and oversight creates major challenges for founders. Reports have found that as many as 80% of workers use unvetted and unapproved AI tools on the job. This isn’t just among lower-level employees. Senior managers and executives often have even higher rates of unapproved AI usage. The problem? Unvetted AI tools often use open-source components which can house major security flaws. The flow of information to and from micro-services, LLMs and database servers can be difficult to track, with the potential for serious connections and permissions vulnerabilities. The Vercel breach exposed a huge amount of database credentials, API keys and third-party integrations. This happened simply because an AI tool was given permission to read software environment variables. In some cases, cyberattackers, who insert false or misleading information into the training data, intentionally “poison” public machine learning models. This can make the AI malfunction in ways that trigger it to provide wrong answers, leak sensitive information or behave in a biased way, even when the model seems to be functioning normally. As agentic AI becomes more widely used, the risks grow exponentially. Agentic AI’s capabilities to carry out complex series of tasks without oversight can be a boon for time-strapped founders. It also allows AI agents to be used for increasingly sophisticated and devastating attacks if they are compromised. Minimize the risks, and maximize the results. For founders, the same risks that exist from a “standard” cyberattack also exist within the AI software supply chain, but at scale — potential regulatory, legal and financial accountability, significant downtime, and lost trust. All of those become even greater risks when founders don’t do their due diligence on their entire AI software supply chain. So how do you minimize the risks for your organization? Start by comprehensively vetting tools your organization uses. Even basic steps such as reviewing terms of use and understanding how an AI tool may use data you feed into its system can help reduce risk. For businesses operating in privacy-focused industries, tools should also meet all relevant regulations. You should also carefully vet the developers behind these AI tools. Ask yourself: Has your AI developers regularly updated AI tools, especially for security? Does the developer provide testing and validation results? Are they transparent with how your data is used or stored? What is their reputation like? Even if a given AI system is deemed safe unto itself, it’s important to map out the downstream connections it has with your apps and servers. This ensures that you are managing all relevant identities and workflows safely. The best AI supply chain security strategies take all of these components into account. Securing AI workflows and access When you integrate AI into your stack, you should also adopt many of the same security practices you use with human employees. Zero trust governance, with strict access and authentication controls, can ensure that AI tools only access the information they need to perform their critical functions. Finally, you need to establish clear AI policies and make sure your entire team follows them, on the user’s end as well as the developer’s end. The 2025 Verizon Data Breach Investigations Report found that roughly 60% of breaches had a human element, usually employee error. Ongoing training regarding safe AI use will hopefully keep your team from using unauthorized AI tools that could compromise your systems. Do you know where your AI comes from? As AI adoption accelerates, ensuring it doesn’t compromise your security becomes an increasingly high-stakes game for founders. It’s great to unlock exciting automations and boost productivity. But you need to reduce risk wherever possible. A proactive and informed approach to your AI software supply chain will help you avoid becoming another cautionary tale. EXPERT OPINION BY HEATHER WILDE RENZE @HEATHRIEL

He Spent 18 Years as a Software Engineer. AI Replaced Him in Weeks—and Exposed the Reskilling Myth

For nearly two decades, David was the guy you called when the system crashed at 3 a.m. With 18 years of experience as a senior software engineer, he had seen the industry move from physical servers to the cloud, surviving bubble bursts and economic downturns. He assumed his deep expertise was his ultimate insurance policy. Then the email arrived. It was the standard “restructuring” notice that has become the soundtrack of the modern tech sector. But this time, the chairs weren’t just being rearranged. David’s entire department was being dismantled to make room for a new team of AI specialists. David wasn’t worried at first. He had spent his career learning new languages and frameworks. He bought the books, took the online courses, and prepared to “reskill” into the AI-driven future that every LinkedIn influencer and CEO was shouting about. The reality, however, was a cold shower. After applying to over 100 jobs, the feedback was consistently the same. Hiring managers told him his skills were obsolete. The very experience he viewed as an asset was now being treated as a liability, a relic of a pre-generative era. Today, David doesn’t spend his nights debugging code. He spends them working the night shift as a waiter at McDonald’s. His story is not an outlier. It exposes the reskilling lie. The myth of the easy pivot We have been told a comforting story about the AI revolution. The narrative suggests that while some jobs will disappear, a vast ocean of new roles will open up for those willing to learn. It sounds logical in a keynote presentation, but it falls apart on the ground. Midcareer professionals are being sold a bill of goods. The idea that a 45-year-old engineer with a mortgage and family can simply take a six-week boot camp and compete with 22-year-old AI natives is a fantasy. It ignores the structural ageism and economic realities of the hiring market. Companies aren’t looking for veterans who have “reskilled.” They are looking for specialists who have lived and breathed neural networks for their entire academic lives. The “pivot” is more of a leap across a widening chasm, and for many, the landing isn’t there. Part of the problem is they haven’t yet grasped the disruption confidence cycle that takes place every time a new technology comes along and changes the game. When experience becomes a liability In the traditional business world, 18 years of experience commands a premium. It represents depth of knowledge and a history of successful projects. In the AI-first world, that same history is often viewed as “legacy baggage.” Hiring managers are increasingly biased toward candidates who don’t have “pre-AI” habits. They want people who think in prompts, not in procedural logic. This creates a trap for midcareer professionals who find themselves overqualified for entry-level AI roles but “underskilled” for senior ones. The result is a talent graveyard. Brilliant minds are being discarded not because they can’t learn, but because the corporate machine doesn’t want to pay for the transition time. It is cheaper to hire a specialist than to wait for a veteran to adapt. The leadership failure This is not just a technology problem. It is a leadership failure. Executives are prioritizing short-term AI integration over long-term talent retention. By replacing veteran staff with specialized newcomers, they are hollowing out the institutional memory of their organizations. When a senior engineer leaves, they take more than just their coding skills. They take the knowledge of why certain decisions were made five years ago. They take the understanding of the client’s deep-seated needs. AI cannot replicate that context yet. Leaders who buy into the replace-and-reskill narrative are often surprised by the problems it creates. The new AI systems might be faster, but the loss of human oversight leads to hallucinations that can cost millions. Replacing a seasoned pro with a prompt engineer is a gamble that rarely pays off the way the spreadsheets predict. Seeking alternative paths For those caught in this transition, the traditional job market can feel like a closed door. This is why many are looking toward independent tools and solo ventures to reclaim their agency. Instead of waiting for a hiring manager to validate their AI skills, they are building their own workflows. Some have turned to platforms that simplify complex industries such as travel planning, financial management, legal paperwork, real estate, health care navigation, online education, and small-business operations. It allows them to leverage their organizational skills without needing a corporate badge to prove their worth. The goal for many is no longer about finding a new desk in a glass building. It is about creating a career moat that AI cannot easily bridge. This requires a move from being a worker to being a builder, regardless of the industry. The hidden cost of displacement The human cost of this displacement is staggering. When a senior professional is forced into low-wage service work, it isn’t just a loss of income. It is a loss of identity and social utility. The psychological toll of being told you are “obsolete” after two decades of high-level performance is profound. This creates a ripple effect throughout the economy. Families lose their stability, and the community loses the tax base of high-earning professionals. The “reskilling” narrative acts as a convenient shield for companies to avoid the moral and economic responsibility of their hiring decisions. Redefining the promise If the reskilling lie is to be corrected, the conversation must change. We need to stop telling people that a pivot is easy and start talking about how to protect and adapt existing expertise. We need policies that incentivize companies to retain and transition their veteran staff. True reskilling requires time, investment, and a willingness to value experience. It isn’t a weekend workshop. It is a collaborative process between the individual and the organization. Without that partnership, the “pivot” will continue to be a cliff for many. For the Davids of the world, the McDonald’s shift isn’t a failure of effort. It is a failure of a system that promised a ladder and then pulled it away once he reached the middle. The AI era doesn’t have to be a zero-sum game between veterans and newcomers. But as long as we pretend that “learning to code” or “learning to prompt” is a magic bullet for a 45-year-old, we are setting millions up for disappointment. Leaders must be honest about the limitations of reskilling. They must acknowledge that experience still matters, even in an automated world. Until that happens, the night shift at McDonald’s will continue to be the unintended destination for some of our brightest minds. EXPERT OPINION BY JOEL COMM, AUTHOR AND SPEAKER @JOELCOMM

These 40 Jobs May Be Replaced by AI. These 40 Probably Won’t

A new study measuring the use of generative artificial intelligence in different professions has just gone public, and its main message to people working in some fields is harsh. It suggests translators, historians, text writers, sales representatives, and customer service agents might want to consider new careers as pile driver or dredge operators, railroad track layers, hardwood floor sanders, or maids — if, that is, they want to lower the threat of AI apps pushing them out of their current jobs. Why should anyone heed yet another of the myriad, sometimes conflicting reports in AI’s potential impacts on jobs? Because the researchers behind the new findings really know what they’re talking about. They all work for tech giant Microsoft, which is developing Copilot and related AI apps examined in the study. And those tools, the authors say, risk putting ticket agents and telemarketers out of work far sooner than orderlies and paving equipment operators. The Microsoft study comes as debate continues about the employment threats AI may pose to millions of people in clerical, administrative, communications, marketing, and other jobs. Executives of several tech companies, including AI developer Anthropic CEO Dario Amodei, have alternatively cheered or warned about bots automating a wide range of work tasks, potentially eliminating up to half of all white-collar and entry-level jobs in the process. Other business leaders, notably serial entrepreneur Mark Cuban, believe the tech will generate even more new positions than it erases by assuming a lot of repetitive drudge work. Microsoft’s new research doesn’t offer an opinion on the quantitative consequences on employment that AI will ultimately have. But it does provide clear indication of which 40 jobs are already using apps most frequently — and the contrasting 40 professions reflecting the tech’s lowest levels of penetration. The full text of their findings and the two rankings are available here. These 40 jobs are most threatened by AI Interpreters and Translators Historians Passenger Attendants Sales Representatives of Services Writers and Authors Customer Service Representatives CNC Tool Programmers Telephone Operators Ticket Agents and Travel Clerks Broadcast Announcers and Radio DJs Brokerage Clerks Farm and Home Management Educators Telemarketers Concierges Political Scientists News Analysts, Reporters, Journalists Mathematicians Technical Writers Proofreaders and Copy Markers Hosts and Hostesses Editors Business Teachers, Postsecondary Public Relations Specialists Demonstrators and Product Promoters Advertising Sales Agents New Accounts Clerks Statistical Assistants Counter and Rental Clerks Data Scientists Personal Financial Advisers Archivists Economics Teachers, Postsecondary Web Developers Management Analysts Geographers Models Market Research Analysts Public Safety Telecommunicators Switchboard Operators Library Science Teachers These 40 jobs are least threatened by AI Phlebotomists Nursing Assistants Hazardous Materials Removal Workers Helpers–Painters, Plasterers Embalmers Plant and System Operators, All Other Oral and Maxillofacial Surgeons Automotive Glass Installers and Repairers Ship Engineers Tire Repairers and Changers Prosthodontists Helpers–Production Workers Highway Maintenance Workers Medical Equipment Preparers Packaging and Filling Machine Operators Machine Feeders and Offbearers Dishwashers Cement Masons and Concrete Finishers Supervisors of Firefighters Industrial Truck and Tractor Operators Ophthalmic Medical Technicians Massage Therapists Surgical Assistants Tire Builders Helpers–Roofers Gas Compressor and Gas Pumping Station Operators Roofers Roustabouts, Oil and Gas Maids and Housekeeping Cleaners Paving, Surfacing, and Tamping Equipment Operators Logging Equipment Operators Motorboat Operators Orderlies Floor Sanders and Finishers Pile Driver Operators Rail-Track Laying and Maintenance Equipment Operators Foundry Mold and Coremakers Water Treatment Plant and System Operators Bridge and Lock Tenders Dredge Operators The results were obtained by analyzing 200,000 “conversations between users and Microsoft Bing Copilot.” Researchers then matched those with “measurements of task success and scope of impact, [to] compute an AI applicability score for each occupation.” The jobs with the highest use rates tended to be office positions or other work communicating data or thoughts for specific business purposes. “We find the highest AI applicability scores for knowledge work occupation groups such as computer and mathematical, and office and administrative support, as well as occupations such as sales whose work activities involve providing and communicating information,” the study says. “Additionally, we characterize the types of work activities performed most successfully, how wage and education correlate with AI applicability, and how real-world usage compares to predictions of occupational AI impact.” The upshot of that is data showing the work of ticketing agents, proofreaders, and PR specialists is already being automated at far higher levels than labor provided by housepainters and plasterers, embalmers, ship engineers, and phlebotomists — the technicians who draw blood for medical tests. But despite the study establishing a de facto ranking of the jobs threatened the most — and least — by AI, its authors ultimately waffle a bit on just how big the tech’s impact on overall employment and workplace stability might be. That’s probably not surprising, given they all work for the same Microsoft employer whose business future will largely depend on successfully developing and selling those work-automating apps to other companies. And some of the study’s disclaimers suggest underlining AI’s potential for possibly eviscerating current employee counts wasn’t considered the best messaging for broadening the appeal of apps to prospective customers. “It is tempting to conclude that occupations with high overlap will experience job loss,” they write in one of those hedges on AI’s likely impact on employment. “This would be a mistake, as our data do not include the downstream business impacts of new technology, which are very hard to predict.” On the one hand, authors do specify that AI tools are being used most often for communications tasks like language interpretation, emailing, and composing marketing materials. But on the other, they hasten to add it isn’t clear apps are also being asked to assume the complete array of tasks those workers perform on the job — or whether they’d even be capable of doing that. The researchers similarly note the different objectives employees studied had in using bots. Some of those workers asked apps to entirely handle and complete certain job tasks. But in many other cases, people queried AI assistants about the most effective ways to fulfill work duties themselves, retaining their own value to employers. Meantime, the study’s authors also seem to balance between the higher productivity objectives — and potentially decreased labor costs — that some employers hope AI will provide and employees’ contrasting fears about their job security. Those diverging focuses, the authors say, won’t generate the zero-sum results many warn of — at least not necessarily. “For example, if AI makes software developers 50 percent more productive, companies could raise their ambitions and hire more developers as they are now getting more output per developer, or hire fewer developers because they can get the same amount done with fewer of them,” they say. “Our data is only about AI usage and we have no data on the downstream impacts of that usage, so we only weigh in on the automation versus augmentation question by separately measuring the tasks that AI performs and assists.” But the study also makes it clear that the jobs least likely to be disrupted initially by increased or dominant use of AI are those involving some mixture of manual activity, use of machines, and interaction with people. That combination leaves nursing assistants, hazardous waste removers, car windshield installers, and medical equipment preparers among the professions with the lowest level of app penetration. But even there, the authors create some wiggle room for eventual employment outcomes. They note their research is based on use of AI that Microsoft developed from large language models (LLMs). More focused apps tailored to individual professions — possibly paired with robotic machines — might still leave many manual jobs vulnerable to the tech’s influence in the future. “Note that our measurement is purely about LLMs,” the authors note. “Other applications of AI could certainly affect occupations involving operating and monitoring machinery, such as truck driving.” Meaning, any historians or brokerage clerks feeling fearful about their work after reading the study might want to rethink any plans about rushing into careers as a roustabout or packaging machine operator. Because those professions, too, may come under the growing influence of specialized AI apps in the not too distant future. BY BRUCE CRUMLEY @BRUCEC_INC

Stop Letting ChatGPT and Other AI Chatbots Train on Your Data. Here’s Why—and How

When you interact with a chatbot, there’s a good chance that everything you say, and every prompt you give, isn’t just used to generate replies to your queries. Nearly every chatbot company on the planet also uses the information you provide to train its AI models. This can leave your privacy—and even your employer’s confidential information—exposed. But you can mitigate these privacy risks by telling chatbots not to use your data for training. Here’s how. What is AI chatbot training? In order for a chatbot to provide knowledgeable and (hopefully) accurate answers, the underlying large language model (LLM) that powers it needs to assimilate a massive amount of information, which it then uses to help answer your questions. This process of information assimilation is known as “training.” The more information an LLM trains on, the more intelligent the LLM, ostensibly, gets. LLMs acquire training data from numerous sources, including public websites, social media platforms, encyclopedias, video-sharing sites like YouTube, and, unfortunately, sometimes even without permission from authors, novelists, artists, musicians, and other creatives. But LLMs also get their training data from you, too. Every time you enter a prompt to give a chatbot information, that information is likely being used by the AI company to further train its models. And that can leave your privacy severely exposed. Why you shouldn’t let AI chatbots train on your data It’s generally a good idea not to allow LLMs to train on your data, especially if, in your interactions with a chatbot, you share a lot of sensitive information about yourself. If you talk to a chatbot about your physical or mental health, your finances, or your relationships, you should know that that data is, by default, usually used by the AI company to further train its LLM, which means your most intimate thoughts, worries, and concerns are becoming part of the model. AI companies say they anonymize the information you provide before using it to train their models—but you really just have to take them at their word. Even if they do anonymize your information, that doesn’t mean a bad actor in the future couldn’t use some technique to link all the prompts about a particular health, relationship, legal, or financial issue back to you. And if you are using an AI chatbot for work, you could be exposing your employer to legal and regulatory risks if the data you feed it contains confidential user or client information. Even if it doesn’t, you could inadvertently give away your employer’s corporate secrets, such as proprietary code or sales data. The chatbot may give you the answers you’re searching for, but it will also use all the data you give it to further train its models—and retain that data as part of itself. How to prevent AI chatbots from training on your data All this means that it’s a very good idea to prohibit a chatbot from training on your data. Doing so will not hinder the quality of the results the chatbot provides to you, but it will ensure, as best as possible, that the data you provide to it won’t be permanently absorbed into the bot’s underlying LLM. The good news is that most reputable chatbots—including the four most popular ones: OpenAI’s ChatGPT, Google’s Gemini, Anthropic’s Claude, and Perplexity AI’s Perplexity—now offer ways you can opt out of having your data used for training. Here’s how to tell the big four chatbots to stop training on your data: ChatGPT: Select your profile to access the chatbot’s settings. Select Data Controls. Select “Improve the model for everyone.” Toggle the “Improve the model for everyone” switch off. Gemini: Go to the Gemini Apps Activity settings page. Select the button that says “On.” From the pop-up, select “Turn off.” Select “Got it” in the confirmation box that appears. Claude: Select your profile to access the chatbot’s settings. Select the privacy menu. Toggle the “Help improve Claude” switch off. Perplexity: Select your profile to access the chatbot’s settings. Select the Preferences menu. Toggle the “AI data retention” switch off. Once you’ve done this, none of the big four AI giants should be able to use the prompts and other information you give their chatbots to further train their LLMs. However, since these firms haven’t provided independent auditors with access to their systems, you have to take the companies’ word that they will stop using your data to train their models. Also note that even if AI companies agree not to use your data to train their models, they may retain information from your chats and other information you provide for legal or regulatory purposes for a set period of time. And even with these anti-training orders in place, it’s still a good idea to thoroughly (and correctly) redact sensitive information from any documents before you upload them to an AI chatbot. To get even more privacy when interacting with popular chatbots, consider using proxies like Apple Intelligence on the iPhone or DuckDuckGo’s Duck.ai, which can help better obscure your digital footprint from AI giants. By Michael Grothaus

AI isn’t actually ‘taking’ your job. Here’s what’s happening instead

AI probably won’t take your job anytime soon. At least not all of it. Concerns about artificial intelligence replacing human workers have simmered over the past year as companies slash headcounts, AI models grow more capable of office work and businesses integrate AI more deeply into their operations. AI was the top reason companies cited for job cuts in April for the second month in a row, the executive outplacement firm Challenger, Gray & Christmas said Thursday. “The anxiety around AI at work is real—from fears of job loss to the pressure to keep up with rapidly evolving technology,” Microsoft wrote in a report about how AI is changing jobs released last week. But the reality of AI in the workplace isn’t so black-and-white, experts say. Companies are using AI to automate certain parts of jobs rather than replace entire positions. Business leaders are figuring out what AI can and can’t do, recalibrating existing jobs around responsibilities that can only be done by a human. And thousands of jobs have been cut in the process, with web infrastructure company Cloudflare and cryptocurrency firm Coinbase among the latest to announce staff cuts. “It’s very few jobs that are actually entirely automated away by the current AI and robotics technology that’s out there,” said Alexis Krivkovich, a senior partner at McKinsey & Company who helps lead the company’s People and Organizational Performance Practice. AI is technically capable of automating 57% of work-related activities, Krivkovich said, citing McKinsey research. But that percentage is spread across “pieces and parts” of various jobs and responsibilities across an organization. Nitin Seth, the cofounder of digital services and consulting firm Incedo, claims his company helps clients boost productivity using AI by at least 20% to 25% without reducing staff at the same scale. That’s because AI only handles certain parts of different roles. “You can’t take one quarter of Lisa, one quarter of Jessica, one quarter of Nitin and one quarter of somebody else and make it one person,” Seth said. The fear that AI will take jobs has disrupted the tech industry the most. Software engineers have increasingly embraced the tech to help write code, with 90% of tech workers using AI in their jobs, according to a September survey from Google’s research arm. Stack Overflow, a popular question-and-answer forum for developers, found that 84% of respondents either use AI tools in the software development process or plan to. But a software engineer’s job involves much more than just coding: It entails reviewing the code, designing systems, troubleshooting problems and deciding what to build. Companies may adjust job titles to reflect that, says Boris Cherny, head of Claude Code at Anthropic. “I think by the end of the year, we’re going to start to see the idea of software engineering go away,” he told CNN in March. He thinks the term “builder” might be a more fitting title as the job expands, and writing lines of code becomes a smaller part of it. Sujata Sridharan, who most recently worked at the fintech firm Bolt and has spent roughly a decade as a software engineer, is one of the many engineers living through that transition. Although she uses AI, her work still requires problem solving and critical thinking, she told CNN over email. The difference is that the execution now involves a mix of writing code and prompting AI. “With AI being used more and more, the skills that are actually required on the job have shifted to, are you able to recognize what is the right code quality? Are you able to problem solve?” she said. That’s not to say AI isn’t contributing to job losses; it just likely isn’t taking over full roles. AI has been cited in more than 49,000 job cuts so far this year, the report from Challenger, Gray & Christmas said. Block, the financial tech company behind Square and Cash App, laid off 40% of its staff this year because AI has allowed it to do more with smaller teams. Coinbase is reducing its staff by about 14% in part because AI is enabling engineers to “ship in days what used to take a team weeks,” its CEO said Tuesday. And Cloudflare said the way the company operates has completely changed, adding that its AI use has increased by more than 600% in the last three months alone. It’s possible there will be “some job disruption on the horizon,” according to Dan Priest, PwC’s US chief AI officer. Still, he said he isn’t seeing mass layoffs at most companies and whole categories of jobs aren’t currently at risk. Most companies haven’t yet adjusted its employee metrics and incentives to fit with how AI is changing work, Microsoft said in its report, which surveyed 20,000 workers using AI across 10 countries. Instead, many are simply grappling with which skills are needed from human workers. And the tech landscape may keep changing as AI models evolve and potentially take on more office tasks. For example, Anthropic on Tuesday announced new AI agents built for financial work, like building pitchbooks and crafting credit memos. “It starts at the bottom, and it keeps going up,” said Umesh Ramakrishnan, cofounder and chief strategy officer at executive search firm Kingsley Gate. “And I don’t know where it stops.” By Lisa Eadicicco

The Coasean Singularity: Why AI Is Ending the Org Chart as We Know It

In 1931, a 20-year-old commerce student at the London School of Economics received a modest traveling scholarship and boarded a ship to the United States. His name was Ronald Coase. He had worn leg braces as a child. Economics had found him almost by accident, in a seminar where a professor unpacked Adam Smith’s theory of the invisible hand. “It was a revelation,” Coase would remember decades later. He had planned to study law. Instead, armed with letters of introduction from a Bank of England contact, he spent a year visiting Ford plants, General Motors factories, and a long list of American businesses, puzzling over a question: Why do firms exist? The question came to him while watching how real industries organized themselves. Some activities stayed inside the company. Others were contracted out. So if markets are as efficient as Adam Smith suggested, why build firms at all? Why hierarchies? Why middle managers? Why conference rooms, reporting lines, and org charts? Coase wrote up his answer in an essay published in November 1937, when he was 26. He titled it “The Nature of the Firm.” The idea was simple: Markets are expensive to use. Friction is everywhere. Finding a spinner takes time. Negotiating the price for every yard of yarn takes effort. Writing a contract for every afternoon of work takes a lawyer. These are transaction costs. Many of them are replaced inside the firm by something blunter and cheaper: authority. A manager tells an employee what to do. The firm expands until the cost of organizing one more transaction internally equals the cost of carrying it out through the market. That is the boundary of the firm. The essay was short. Dry, even. Coase published it in Economica. Fifty-four years later, in 1991, it helped earn him the Nobel Prize in economics. For close to a century, the theory has held. It explained why companies got big during the industrial age (coordination was expensive, so internalize everything), why they got flatter during the internet age (email and ERP systems reduced internal coordination costs), and why outsourcing boomed in the 2000s (India’s IT services made market transactions cheap enough to push work outside the firm). Now, in 2026, Claude and other AI agents are pushing Coase’s logic toward its limit. On the supply side, AI is collapsing coordination costs in ways the internet never could. An agent can draft a contract, source a supplier, negotiate terms, and monitor delivery. That persistent friction of the past is finally dissolving. On the demand side, the comfortable middle of the market is thinning. The place where Applebee’s fed families, where Gap clothed the masses, and where standard software packages served nearly everyone has been hollowing out. What remains is a barbell: the ultra-cheap and algorithmic on one end, the hyper-personal and curated on the other. AI did not create these forces. It is accelerating both at once. Driving coordination costs lower on the supply side. Splintering expectations further on the demand side. And the result is rewiring the firm and rewriting our jobs. Firms will not disappear. Jobs will not disappear. But in a world moving toward near-zero transaction costs, both are being remade. This is what it looks like when Coase meets Claude. I call this the Coasean Singularity. The Death of Normal In 2007, you walked into a theater and bought a ticket. No research. No TikTok. You picked a movie because it started at 7:30 p.m. and your date liked George Clooney. That was enough. Michael Clayton cost $21 million. It was nominated for Best Picture. It was a movie for adults, made by a studio, distributed widely. People who sold insurance watched it. People who taught high school watched it. People who were on their second date and just picked something watched it. That movie does not get made today. What replaced it is a barbell. One end is the $200 million superhero franchise. The other end is the $7 million A24 film about grief and mushrooms. The middle is gone. Those boutique masterpieces are smarter. They reach exactly those who already know what they’re looking for, thanks to Instagram and TikTok. The middle of every major consumer market is collapsing. Films, restaurants, music, retail, careers. What is replacing it is a barbell. Enormous weight on each end. Nothing in the center. Walk into a strip mall and you can see it. In May 2024, Red Lobster filed for bankruptcy with roughly $300 million of debt. James Berke, 23, a server in New Jersey, woke up Monday to the place padlocked. The parking lot was full, and the doors were locked, and nobody had said a word. “They made us work Mother’s Day to get that quick buck, and then they closed us.” Freezers were carted out. Booths were auctioned off. Lobster tanks were drained. More than 130 locations closed. Red Lobster wasn’t alone. TGI Fridays shuttered 185 locations. Denny’s closed up to 150. Applebee’s trimmed 47. Hooters, Buca di Beppo, and dozens of restaurant chains filed for bankruptcy in 2025 alone. But Sweetgreen’s shredded kale and wild rice bowl is doing fine. The taco truck is doing fine. The premium survived. The cheap survived. The place where a janitor and an accountant once sat in the same room and ordered from the same menu is what’s disappearing. And no one seems to think that is a problem, which tells you exactly who the replacements were built for. They are more interesting; they serve fewer people. The cultural monoculture vanished somewhere around the Game of Thrones finale in May 2019. That was the last time 19.3 million people watched the same thing on the same night. Nothing since has come close. Algorithmic feeds now create individual reality tunnels. TikTok spins new ones every few weeks. Shein adds six thousand new styles a day. Trends flash, peak, and vanish. That’s how consumers stopped wanting “good enough for people like me.” They want good enough for me. You can see that inside your own company, too. Microsoft and LinkedIn found that 75 percent of knowledge workers already use AI at work. And 78 percent of AI users are bringing their own tools, because the company-issued version feels generic. Bosses call this shadow IT and treat it as a compliance problem. But these were not rogue employees. They were the diligent ones. The ones who stayed late. They just wanted tools that worked the way they worked. The market has ceased to be a mass. Demand is distributed, specific, shifting, and impatient. People want software, clothes, meals, careers shaped to their exact workflow, their exact body, their exact mood. Companies have taken notice. Pernod Ricard now tracks how AI models talk about its whisky brands and uses those insights to refine its marketing and creative. Instacart built a ChatGPT plugin that lets you add groceries to your shopping cart mid‑conversation, and has since made it a full in‑ChatGPT flow. These are real adaptations. They are also mostly interface moves. They will not be enough. When you get AI to highlight Ballantine’s as affordable in real time, that does not help you launch a limited-edition single malt in seventy-two hours because a TikTok bartender in Seoul just made peat smoke go viral. Serving those ephemeral demands requires a supply structure that can spin up fast, go narrow, and dissolve when the moment passes. Most firms still plan in annual cycles, manufacture in big batches, and launch on timelines measured in quarters. Which brings us back to the 20-year-old in leg braces who sailed to America in 1931. The Supply Side: When Coordination Gets Cheap To understand what is happening on the supply side, I spoke with Sangeet Paul Choudary, author of Reshuffle. I asked him what most companies get wrong about AI. Everyone starts with tasks, he said. A lawyer drafts contracts faster. A marketer generates copy faster. A developer writes code faster. Fine. But that is still the same workflow. The deep shift, he argues, comes from cutting the effort it takes to turn one team’s output into another team’s input. The meeting on your calendar because engineering and marketing speak different languages. The email chain because the CRM does not talk to the ERP. In Coase’s language, those frictions are transaction costs. And AI’s biggest economic payoff comes from driving those transaction costs down to nearly zero. When AI agents can negotiate terms, monitor quality, interpret documents, and reconcile outputs, firms can distribute work more widely, without losing the grip they once needed hierarchy to provide. What does that look like in practice? It starts with what Choudary calls the atomic unit of work: the smallest piece of productive activity that can be separated and then assigned. Take fashion. For Zara, the atomic unit was the collection: a coherent batch of designs that had to fit the brand, be manufacturable, and move fast enough to catch a trend. That required tightly coupled human judgment across design, sourcing, production, and logistics. So Zara built a supply chain around speed and coordination, with production close to Spain and fresh inventory moving out of La Coruña twice a week. Then Shein kept shrinking the unit of work. A collection became a set of fragments. Then the fragments got smaller still, until what remained was a single design decision. “The designer still exists,” Choudary told me. “But their creative bandwidth has been compressed to almost nothing.” Work that once sat with a designer who could hold manufacturing constraints, brand identity, and seasonal timing in mind can now be split into narrow prompts: Reference these three images. Draw a collar. Rework the cuff. Try a different hemline. Human judgment still matters. But it shows up less. More of the intelligence now sits in the system that parcels out the work, recombines the pieces, and decides what deserves to scale. In a cramped Guangzhou workshop, workers would hunch over sewing machines under hard fluorescent light, finish a piece, slide it into a turquoise bag, and toss it onto the pile. The system reacts to each signal as it arrives, from an Instagram click to an Amazon sale. That helps explain why Gap or Under Armour struggle. They are still organized around the older unit of work: long cycles, big production runs, broad seasonal catalogs. Their unit is still too large. You can see it in software, too. Adobe treated the file as the basic unit of design work. You work on the file. You pass the file around. Figma treated the element, the text block, the shared object, as the unit. Suddenly everyone is on the same canvas. Version reconciliation fades. Governance stops being something imposed after the fact and becomes part of the workflow itself. By the time Adobe recognized the threat, it was serious enough that the company attempted a $20 billion acquisition. Regulators blocked it. So the pattern is consistent. The unit of work gets smaller. The pieces become more modular. Coordination gets cheaper. And once that happens, more of what once required a firm begins to look like something that can be orchestrated across a network instead. Where Supply Meets Demand The convergence matters because it changes the minimum viable size of a firm. When coordination gets cheap and demand gets weird, more businesses can exist at smaller scale. Think YouTube channels versus TV shows. It took 150 people to make a TV show. Now 5 to 10 people with AI tools can run a wildly successful channel. The same explosion is coming to software. A software company used to require 10,000 customers, 50 employees, and $1 to $5 million in capital. But AI is threatening to drop those requirements to 500 customers, 2 people, and minimal funding. That is why Wall Street repriced software companies so much. In early February 2026, software and services stocks shed roughly $830 billion in market value over six trading days. Salesforce, Adobe, ServiceNow all get hammered. The market realized that software barriers built on workflow lock-in will dissolve if AI agents can coordinate across tools without anyone agreeing on a standard. None of this means large companies disappear. They never do. When e-commerce arrived, Sears collapsed, but Amazon rose in its place. The form changes. The scale persists. So what does a big company look like after the Coasean Singularity? How Large Organizations Survive (When They Do) The honest answer is that most won’t in their current form. But there is an alternative, and it has a proof of concept. It is called Haier, the world’s No. 1 major-appliances brand for 17 years running. In December 1984, a 35-year-old named Zhang Ruimin was sent to run a failing refrigerator factory in Qingdao. His predecessors had all quit. Annual turnover was 3.48 million yuan. The factory was losing 1.47 million. Workers urinated on the floor. A few months later, a customer returned a faulty refrigerator. Zhang checked the inventory and found 76 defective units, about a fifth of the stock. He lined them up on the factory floor, handed workers sledgehammers, and told them to smash every one. Then he picked up a sledgehammer himself and brought it down on a refrigerator door. Each unit was worth four years’ wages. Some workers wept as they swung. That sledgehammer now sits in Haier Group’s headquarters museum, beside the refrigerators and washing machines that came after it. But the sledgehammer was only the beginning. Over the next two decades, Zhang pulled apart the hierarchy itself, reorganizing more than 80,000 employees into over 4,000 self-managing micro-enterprises. Each one operates like a startup. It owns its own profit and loss. It makes key decisions. It stays tied to a specific user need. Employees do not collect a fixed salary simply for showing up. Instead, their income rises or falls with the value they create for customers. A strong micro-enterprise earns its place and stays; a weak one does not. Zhang saw the traditional enterprise as a ship. One captain, one direction, everyone on board. Fine for calm seas. But the seas were not calm. What he wanted was a rainforest: diverse, distributed, self-organizing, resilient because no single failure could bring the whole system down. So he broke the ship apart and let the pieces find their own currents. One micro-enterprise leader said, “My wife used to complain that I didn’t come home from work until after 9:00 at night. But now she is very patient and proud about the hours I keep, because she knows I am building my own company and working for the benefit of our own family.” He paused. “And I am making my own decisions, not acting on the decisions someone else has made.” When Haier acquired GE Appliances in 2016, GE’s market share in home appliances had languished around 2 percent for four years. Under the micro-enterprise model, it surged tenfold. Revenue more than doubled. Small teams swarmed opportunities the way a traditional hierarchy never could. So why isn’t this just fragmentation? Why isn’t it 4,000 independent contractors with a logo? Because when Haier operates as a platform, a kind of internal market with a shared brand and balance sheet, it does what a contract cannot. It provides accountability. It provides trust. It provides a brand, which in a world of infinite AI-generated noise is one of the few filters customers actually rely on. And it provides the capital to absorb failure. “The task is not to turn Haier’s internal staff into entrepreneurs, but rather to attract all the entrepreneurs in society onto our platform,” Zhang says. Most micro-enterprise bets lose. Only an entity with sufficient scale can sustain the portfolio of experiments required to find the ones that win. The Question That Matters I went back to Coase because I needed an anchor. Transaction costs are falling. The unit of work is shrinking. The middle of the market is thinning. An economy dominated by fragmented supply, splintered demand, and AI agents handling the coordination in between does not resemble anything most of us were trained to navigate. All the while, the forces driving us toward the Coasean Singularity are too theoretically fundamental and too empirically visible to be a mirage. My nephew is fourteen years old, living in New Zealand. When I think about the economy he will inherit, I do not think about which AI tool will be dominant. By the time he enters the workforce, “company” might mean something his father would not recognize. A temporary coordination pattern. A portfolio of bets. And “career” means portfolio. If the micro-enterprise economy works only for the relentless, for the people who can treat career volatility as a lifestyle brand, then our economy is also a tournament. And tournaments produce a few winners and a lot of wreckage. The urgent question is how to position yourself. Still, the defining question is whether we can build institutions and shared norms strong enough to make this new economy socially durable and make it possible to live a decent life. We have not yet passed the Coasean Singularity. Pay close attention to who is shaping the rules, and find a way into the conversation. EXPERT OPINION BY HOWARD YU @HOWARDHYU

Inside OpenAI, This Productivity Hack Is Giving Workers Their Own Chief of Staff. You Can Use It Too

Inside OpenAI, the company behind ChatGPT, employees both technical and non-technical are using Codex, the company’s agentic coding app, to handle an increasing amount of work. Codex is OpenAI’s label for its AI-coding platform, which has been accessible through the cloud for over a year. It experienced a huge surge in growth, however, following the release of a dedicated Codex desktop app for Mac and Windows PCs. Unlike the cloud-based versions of Codex, the desktop app is capable of connecting to a computer’s local filesystem, editing existing files, and creating whole new files. In effect, this means that by using the Codex app, users can direct AI agents to do any work that requires a computer. Within OpenAI, non-technical employees are using it in ambitious ways. “The investor relations team is like three people,” says Alex Embiricos, the product manager responsible for Codex, and has been using Codex to monitor the influx of cash from the company’s recent $122 billion fundraise. During a recent all-hands meeting, Embiricos recalls, CFO Sarah Friar told staff that a member of her team had about half an hour to kill, and in that time “vibe coded a dashboard that just showed the incoming expected transfers lighting up, and sent it to Sarah over the weekend.” Two members of the Codex team spend much of their time hosting office hours and answering questions from non-engineering teams at OpenAI. By monitoring if usage rates spike after one of these sessions or hackathons, the Codex teams can understand more about what use cases certain divisions of the company are discovering for themselves. Laura Peng, one of the members of the Codex product operations team, says that many employees are getting utility out of simple dashboards that connect to their email and work communications service like Teams, Slack, and Google Meet, along with data sources like Excel and Sheets. Employees are monitoring emails, keeping track of deadlines, and summarizing and responding to Slack messages. “I feel like I almost have my own chief of staff, in a way,” says Peng, “just making sure that I’m staying on top of my job.” As OpenAI’s ever-growing number of employees get up to speed with Codex, Peng says, “the floor has just [been] raised for everyone, where people’s level of curiosity about a tool they previously would have been really intimidated by has just grown.” Personally, Peng is currently planning a trip to Korea, and used Codex to create an interactive interface where she could “kind of click on the different cities that we were visiting and see how far everything was from each other.” One popular method for learning how to get the most out of Codex, says Peng, is developing small games. Maybe you start by making a classic Snake game, she suggests, and then ask Codex to make the snake neon pink, and then turn the targets that the snake eats into digital strawberries. This kind of rapid iteration is fantastic for teaching people about the open-ended problems that Codex can solve. Katy Shi, a research lead on Codex, also creates games with Codex, but usually as a means of benchmarking the agent when new updates or models are released. One popular challenge is to see how Codex reacts when being tasked to make a first-person version of Tetris from the perspective of the piece. Part of Shi’s work involves crafting the personality of Codex, and finding a way to thread the line of being helpful without becoming too trigger-happy or overeager in its actions. To be sure, OpenAI isn’t the only company that is bringing the benefits of agentic coding to non-engineering work. In January, the company’s chief rival Anthropic released Claude Cowork, a feature on the company’s Claude desktop app that takes some of the abilities of its popular Claude Code product, and packages them in a more beginner-friendly user interface. Instead of creating two separate products for engineering and knowledge work, OpenAI has elected to make Codex adept at handling both types of tasks. Recently, OpenAI has begun adding some aspects of Codex back into ChatGPT, such as Workspace Agents in ChatGPT, which take the form of little animated characters that workers can assign to handle specific tasks, and then share among their organization. Eventually, the plan is for Codex and ChatGPT to merge into a single “super-app” that handles everything for you in a simple interface. One-off features like Workspace Agents are just a taste of what’s to come in the future. For now, Embiricos is laser-focused on making the Codex desktop app as good as it can possibly be, and those efforts are paying off. Since the February 2 launch of the Codex desktop app, OpenAI says, the app has grown to over 4 million weekly active users. At some point, Embiricos says, work will shift over to making the mobile version of Codex as useful as the desktop version, and then the need for desktops and laptops will eventually just go away. “I don’t even want to have to open my computer,” says Embiricos. It won’t be long before everyone in the world has a true “personal assistant on their phone.” BY BEN SHERRY @BENLUCASSHERRY

Google’s New Workspace Intelligence Is About to Be Your Ultimate AI Co-Worker

Some of the most popular work-focused apps of all time are about to become more intertwined than before, thanks to artificial intelligence. Google has announced Workspace Intelligence, a new semantic layer for its Google Workspace suite of products. The new capability enables Gemini-powered AI agents to understand and replicate the unique context of your workplace and job when you use products like Google Sheets and Google Docs. Yulie Kwon Kim, Google Workspace’s head of product, says that Workspace Intelligence is intended to emulate the institutional knowledge of a long-tenured employee, but for AI systems. “There are a lot of LLMs out there that are powerful and capable,” says Kim, “but they end up being generic. They don’t know your institutional history, how you like to format your professional voice and preferences.” The key to creating an actually-useful AI co-worker, says Kim, is to provide AI models with the “rich organizational context” found in a company’s documents, emails, presentations, and messages. Because Workspace Intelligence automatically scans your emails and messages, it can generate content that sounds as if it were written by you, and can transform its outputs to replicate your unique voice. With this context, Gemini will know to format a spreadsheet in Google Sheets the right way, for example, or know to never include exclamation points in your emails. Or maybe to include them all the time! With Workspace Intelligence, Kim says that conversations with Gemini will become more like running a “command center for your work,” enabling users to pull in data from sources including Google Drive, Gmail, and Sheets. Google says that the chat feature, now named “Ask Gemini in Chat,” will make it easier for workers to find files, schedule meetings with coworkers, and get recommendations for the first actions they should take after opening their laptop. In Sheets, Google’s web-based spreadsheet software, users can harness Workspace Intelligence to develop and edit spreadsheets based on data from Gmail or Google Drive. The tech has also enabled a new feature called Sheets Canvas, which enables Gemini to turn spreadsheets into dynamic mini-apps and dashboards. In Docs, Google’s cloud-based word processor, Workspace Intelligence can be used to add infographics to text documents, and to edit documents based on comments left by the user. To be sure, Google has already offered AI tools in different forms in Workspace. Over the past few years, Google has added windows for interacting with Gemini to Docs, Sheets, and Slides, but until now, those interactions didn’t have access to an organization’s data. Workspace’s apps currently include a button that brings up a Gemini chatbot that can make edits to your files, but until now, context from your larger organization wouldn’t be automatically pulled in. Workspace Intelligence is just one of many announcements that Google is making at Google Cloud NEXT, its annual conference in Las Vegas, and Kim says she used the semantic layer to prepare for her trip. To manage the “overwhelming” amount of slide decks, chats, emails, and docs getting thrown her way, Kim says, she has used Ask Gemini in Chat to develop briefings for all the customer meetings she has scheduled, along with separate briefings for her onsite internal team meetings. Workspace Intelligence is now available for all paid Google Workspace users. BY BEN SHERRY @BENLUCASSHERRY

Duolingo’s AI U-Turn Is a Warning for Other Companies

At many big companies these days, finding ways to use AI to do your job better isn’t a suggestion. It’s a requirement. As The Wall Street Journal recently reported, “From small startups to giants including Amazon.com, Alphabet, Google, and Meta Platforms, tech companies are measuring [AI use] with an eye on productivity gains and in certain cases factoring it into performance reviews.” Given the industry’s mad dash to realize the potential productivity gains of AI and keep ahead of the competition, leaders’ desperation to have employees embrace AI makes sense. But is tracking and scoring AI usage in performance reviews the best way to go about it? The experience of learning app Duolingo, as well as some fascinating recent research, suggests companies should think carefully about how they evaluate employees’ AI. The potential for unpleasant and unintended consequences is high. Duolingo’s AI U-turn Duolingo embraced AI early and enthusiastically, stirring controversy. So on a recent episode of the Silicon Valley Girl podcast, host Marina Mogilko wanted to dig into the details of the company’s AI push. She asked CEO Luis von Ahn to explain how Duolingo tracks and evaluates AI use as part of the performance review process. But von Ahn pushed back against the premise of the question. “For a while, it was part of performance reviews. We decided not to do that,” he clarified. Why the change of heart? “I sent a memo to the company that said, ‘Part of your performance review is going to be usage of AI.’ And we found that people were … kind of asking, ‘Do you just want us to use AI for AI’s sake?’” he explained. The focus on maximizing AI use over maximizing AI benefits wasn’t what Duolingo was after. Von Ahn changed course. “We said, ‘No, look, the most important thing in your performance is that you are doing whatever your job is as well as possible.’ A lot of times AI can help you with that, but if it can’t, I’m not going to force you to do that,” he said. “We backtracked from that because it felt like, rather than being held accountable for the actual outcome, we’re trying to just push something that in some cases did not fit.” Beware workslop Duolingo discovered that forced, performative AI use wasn’t actually benefiting anyone. Instead, it was creating AI showpieces to cite when performance review season rolled around again, and crowding out other, more impactful work in the process. Power to management for recognizing the problem and reversing course. But is this only the unique experience of one particular company? Or are other leaders likely to discover, as von Ahn did, that forcing AI usage creates time-wasting, resource-consuming distraction? Recent research from Stanford University and coaching platform BetterUp suggests the problems that cropped up at Duolingo are a danger that more managers need to consider. And they gave that danger a catchy name: workslop. You may have heard the word because it ricocheted around the internet once the researchers coined it. That instant popularity probably reflects how many of us recognized the widespread problem it describes—low-quality, AI-assisted output that forces others to spend time understanding, processing, and fixing it. Just how widespread is workslop? In an initial study, the researchers crunched some numbers and came up with a startling estimate. “Employees reported spending an average of one hour and 56 minutes dealing with each instance of workslop. Based on participants’ estimates of time spent, as well as on their self-reported salary, we find that these workslop incidents carry an invisible tax of $186 per month. For an organization of 10,000 workers, given the estimated prevalence of workslop (41 percent), this yields more than $9 million per year in lost productivity,” the researchers wrote on HBR. How Duolingo accidentally encouraged workslop Using AI to cut cognitive corners and/or impress the boss costs companies millions a year. It also annoys workers tremendously. And leaders, the researchers discovered in a subsequent study, are often guilty of accidentally making the problem worse with AI mandates like the one originally instituted at Duolingo. “Many leaders are facing pressure to make responsible investment decisions about AI in the face of uncertainty and macroeconomic pressures,” the researchers wrote in a second HBR article. “In response, leaders are using a blunt strategy, mandating that employees use AI broadly and quickly.” The predictable result of these less-than-well-thought-out AI mandates isn’t tech-driven productivity gains. It’s more workslop, more wasted time, and more frustrated employees. Better ways to get employees to use AI Bosses thinking of following the lead of tech giants like Meta and using brute force to compel teams to use AI more should take Duolingo’s experience as a warning. Everyone agrees that AI will ultimately have huge upsides for businesses. The stakes are high, and pressure is on leadership. But rushing out blanket AI mandates has serious downsides. So what should leaders do instead of one day announcing to workers that they’ll be evaluated on their AI use at their next performance review and hoping for the best? In their second HBR article, the researchers lay out a handful of suggestions. They include creating an atmosphere of trust where people can discuss their AI experiments honestly, warts and all, and investing in training and knowledge-sharing initiatives between employees. Some companies might even consider creating a position of “AI collaboration architect” to help employees figure out the best ways to deploy AI. EXPERT OPINION BY JESSICA STILLMAN @ENTRYLEVELREBEL